Building an AI Second Brain with Astra

Watch the full walkthrough on YouTube

You have probably seen the AI second-brain posts.

Beautiful Obsidian knowledge graphs. Hundreds of linked notes. Somebody telling you that you really must install this fancy Github repo or you are falling behind.

Then you try to build an AI brain and spend the afternoon learning all new proper nouns.

Here is the less impressive reality:

It is a folder

That’s sorta it. Sorry.

An AI second brain is a folder containing useful context about you. Your work, projects, goals, preferences and decisions you have already made.

Inside are ordinary text files with names such as about-me.md, projects.md and how-I-work.md. Markdown is formatted text. You can open it, read it, edit it and keep it.

ChatGPT, Claude and other models can read the same current notes which means you have a brain that operates all your tools. That’s the crux here.

It keeps the decisions, outcomes and useful context worth carrying forward. Every new chat can start from there instead of zero.

AND the more you use the AI brain the more knowledge it acquires about you, how you work, your business and life etc. etc. It gets better with use!

Get AI to build it

Alright let’s build it.

Technical right? Nah. You know me.

The main thing here is get the Desktop app. For Claude or ChatGPT, doesn’t matter. I use ChatGPT but both are excellent.

We need to use the desktop app because we’re building our AI brain locally on our computer. It doesn’t “live” on ChatGPT.

Open the ChatGPT desktop app, use Codex with Astra, and say:

I want a simple folder you can use to understand me and my work. Interview me first, explain everything in normal language and organise the files for me.

Then talk. Use a microphone so you can blab without editing yourself into a corporate mission statement.

Ask it to interview you one question at a time and follow each answer three or four levels deep. What do you do? What are you trying to change? What keeps eating your time? What have you tried already? How do you make decisions? What do you keep avoiding?

You can add a handful of useful documents, screenshots and websites. A smaller collection of good context beats dumping an entire hard drive into the folder so be parsimonious here.

Read the notes back. Correct anything that sounds wrong. Keep talking until you recognise yourself in the files.

The AI will then take everything you said and build your AI brain. It’ll work out the best structure, what folders to include, what file to add. It’ll write up the contents of those files. It’ll do everything for you.

There is zero need for you to go and buy an AI Brain template or expensive piece of software because guess what? Your AI brain is different to everyone’s elses’. Because your life is different. Your business. Your work. Your priorities. Totally unique and individual. So…using someone else’s structure makes zero sense.

So we use the above process instead. That’s it. You’ve got the basics of your AI Brain. Congratulations. It was pretty easy eh?

Give it one useful job

OK so we have a nice AI brain folder. Cool.

But…it’s not doing much yet.

We need to connect it to tools.

Go into Codex and Claude and literally say “hey my AI Brain lives here” and give it the location. Then instruct it to refer to the AI brain for its tasks for additional context.

And to keep notes about the work it’s been doing in the brain. This allows for other chats and indeed other AI tools to also pick up on the goings ons across your whole ecosystem.

Now pick one real task you repeat: prepare the Friday report, plan next week's content, review a proposal or turn a customer call into the follow-up work. Let the AI read the relevant notes before it starts. And let it rip.

This is how the work starts reflecting your actual business.

If you want the next steps check out the full Youtube video walking you through but for now just get the foundational step done!

My earlier guide to building one shared AI vault goes deeper on sharing context between models. The GitHub beginner's guide covers version history, and the Codex personal-assistant walkthrough shows how to put that context to work.

To the Task,

Kyle